reflow_taskpacks

Reusable Reflow graph packages for higher-level workflows — object detection, pose estimation, hand/face landmarks, media transcoding.

Most users should depend on reflow_rt with the ml feature, which re-exports this crate as reflow_rt::taskpacks. Direct use is appropriate for authoring new taskpacks or shipping them as standalone packages.

What a taskpack is

A taskpack is a serialized GraphExport (plus metadata) that wires together preprocess, inference, and decode actors into a reusable subgraph. You can embed it into a parent graph with SubgraphActor and expose only the ports your app touches.

Because taskpacks are just graph data, they:

• Can be versioned, downloaded, and swapped at runtime.
• Get validated against the standard template registry.
• Require no framework-level hardcoded pipelines.

Relationship to other crates

• Graph shape comes from reflow_graph.
• Actors inside taskpacks come from reflow_cv_ops, reflow_ml_ops, and reflow_components.
• Executed by reflow_network (or through reflow_rt).

License

MIT OR Apache-2.0.